The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	Extracting product development intelligence from web reviews
Author:	Yagci, Ismail Artun
View Online:	njit-etd2014-022 (xvi, 209 pages ~ 4.4 MB pdf)
Department:	Department of Mechanical and Industrial Engineering
Degree:	Doctor of Philosophy
Program:	Industrial Engineering
Document Type:	Dissertation
Advisory Committee:	Das, Sanchoy K. (Committee chair) Bladikas, Athanassios K. (Committee member) Passerini, Katia (Committee member) Gopalakrishnan, Shanthi (Committee member) Cai, Wenbo (Committee member)
Date:	2014-01
Keywords:	Product development Opinion mining Web reviews Product design Design methods Online product reviews
Availability:	Unrestricted
Abstract:	Product development managers are constantly challenged to learn what the consumer product experience really is, and to learn specifically how the product is performing in the field. Traditionally, they have utilized methods such as prototype testing, customer quality monitoring instruments, field testing methods with sample customers, and independent assessment companies. These methods are limited in that (i) the number of customer evaluations is small, and (ii) the methods are driven by a restrictive structured format. Today the web has created a new source of product intelligence; these are unsolicited reviews from actual product users that are posted across hundreds of websites. The basic hypothesis of this research is that web reviews contain significant amount of information that is of value to the product design community. This research developed the DFOC (Design - Feature - Opinion - Cause Relationship) method for integrating the evaluation of unstructured web reviews into the structured product design process. The key data element in this research is a Web review and its associated opinion polarity (positive, negative, or neutral). Hundreds of Web reviews are collected to form a review database representing a population of customers. The DFOC method (a) identifies a set of design features that are of interest to the product design community, (b) mines the Web review database to identify which features are of significance to customer evaluations, (c) extracts and estimates the sentiment or opinion of the set of significant features, and (d) identifies the likely cause of the customer opinion. To support the DFOC method we develop an association rule based opinion mining procedure for capturing and extracting noun-verb-adjective relationships in the Web review database. This procedure exploits existing opinion mining methods to deconstruct the Web reviews and capture feature-opinion pair polarity. A Design Level Information Quality (DLIQ) measure which evaluates three components (a) Content (b) Complexity and (c) Relevancy is introduced. DLIQ is indicative of the content, complexity and relevancy of the design contextual information that can be extracted from an analysis of Web reviews for a given product. Application of this measure confirms the hypothesis that significant levels of quality design information can be efficiently extracted from Web reviews for a wide variety of product types. Application of the DFOC method and the DLIQ measure to a wide variety of product classes (electronic, automobile, service domain) is demonstrated. Specifically Web review databases for ten products/services are created from real data. Validation occurs by analyzing and presenting the extracted product design information. Examples of extracted features and feature-cause associations for negative polarity opinions are shown along with the observed significance.